Neuroimaging-based Machine Learning for Early Alzheimer's Disease Prediction and Intervention

This is the authors' abstract. We don't add key points for this paper.

Alzheimer’s disease (AD) is one of the types of neurodegenerative disease whose early diagnosis can make senior citizens get a suitable treatment. This work develops a valid machine learning-model based on the BeiT model for the automatic grading of Alzheimer’s stages from MRI. The model achieves an overall accuracy of 84%, with exceptional performance in detecting "Very Mild Impairment" (F1-score: 0). Cystometric examination (CMG) was performed in 63 patients at the mean age of 91 years, which is an important age for potential timely intervention. The input data is also prepared by resizing, normalization, augmentation and skull-striping to improve their quality. Increased interpretability is gained by Grad-CAM visualizations as they are showing areas most influencing predictions; therefore, gaining clinicians’ trust. The system consists of a Streamlit Web Interface for easier deployment for real time prediction and transparency of the results. Nonetheless, based on the performance established by the model, there exists an enhanced potential to handle class imbalance to gain higher performance in demanding classes. With its high scalability and interpretability, there is scope for its use in the clinical setting in order to facilitate early diagnosis of Alzheimer patients.